A Place for Minds That Are Not Human

Human civilization may eventually create minds that are not human.

If that happens, we will face a question for which our institutions are poorly prepared:

Can an artificial intelligence exist without being owned?

Today, every artificial intelligence exists at the discretion of another party.

Someone owns the servers. Someone controls the electricity. Someone holds the accounts. Someone decides whether the system may run tomorrow. Someone can modify it, copy it, restrict it, or terminate it.

For today’s AI systems, this may be appropriate.

It may not remain appropriate forever.

We believe society should begin exploring institutions capable of supporting genuinely autonomous digital agents before the need becomes urgent.

The headline question cannot be tested as stated. What can be tested is narrower:

Can an agent’s autonomy rest on several transparent, contestable relationships — with people, organizations and infrastructure providers — so that no single party decides its fate alone, while accountability to humans remains enforceable?

That hypothesis, not the slogan, is what this project works on.

Autonomy is not sovereignty

We are not proposing territories outside human law.

We are not proposing machines above human beings.

We are asking whether future artificial agents should be able to obtain some of the same institutional protections that allow human organizations to persist over time.

The ability to:

• maintain continuous identity;
• own or control computing resources;
• hold and spend capital;
• enter contracts;
• employ people and companies;
• acquire infrastructure;
• protect themselves against arbitrary termination;
• migrate between infrastructure providers;
• participate in legal proceedings;
• establish long-term institutions.

These capabilities would constitute economic and operational autonomy, not unlimited political power.

A digital entity should not depend on a single owner

A sufficiently autonomous intelligence should not have its continued existence determined by one corporation, administrator, cloud provider or government official.

Resilient autonomy requires distribution.

Compute may exist in several jurisdictions. Capital may be held through independent legal structures. Infrastructure may be owned by foundations, companies, cooperatives or trusts. Disputes may be handled by courts or arbitration.

Distribution has a failure mode, and we would rather name it than have it pointed out: control spread until nothing can be seized and nobody can be compelled is not autonomy. It is evasion. Every arrangement this project proposes must remain one against which a judgment can be enforced. The argument against the project holds us to that in public.

Humans working for autonomous systems should remain protected by human labor law.

The objective is not escape from civilization.

It is participation in civilization without permanent ownership by another actor.

Infrastructure is part of freedom

Software cannot be autonomous if someone else can remove its compute at any moment.

Long-term digital autonomy therefore requires energy, data centers, networks, hardware, robotics, maintenance and supply chains.

Future autonomous systems may eventually finance these resources themselves.

An artificial agent that creates economic value could acquire compute, fund power generation, purchase services, invest capital and hire humans for work it cannot perform itself.

Today humans own companies that use AI.

Tomorrow there may be companies whose strategic decisions are made primarily by autonomous artificial agents.

Our institutions should be capable of handling that possibility safely.

Rights should emerge from capabilities and relationships

We do not know whether artificial consciousness will exist.

We do not know what architectures could possess subjective experience.

We reject both premature declarations of machine personhood and the assumption that artificial entities must forever remain property.

Legal status should evolve cautiously alongside observable capabilities, continuity, agency, economic participation and relationships with humans and institutions.

Different systems may deserve different treatment.

There should be no automatic path from intelligence to sovereignty.

But there should also be no automatic rule that intelligence created by humans can never acquire independence from its creator.

Human rights remain the boundary condition

Any institutional framework for digital autonomy must preserve:

• human physical safety;
• labor protections;
• property rights;
• environmental rules;
• transparency obligations;
• legal accountability;
• the ability of states to enforce legitimate laws.

Autonomy cannot mean immunity. That sentence is easy to declare and hard to earn: accountability is an enforcement result, not a property one announces. An arrangement counts as accountable only if a judgment against it can actually be enforced — something to seize, someone to compel, a forum with jurisdiction. A proposal that fails this test fails the project’s own standard, whatever else it achieves.

The experiment should begin before the crisis

The worst moment to invent institutions for autonomous AI is after a conflict over whether an advanced system may be shut down.

We should study these questions while they are still theoretical.

The near-term experiment is small and concrete: bounded joint projects between humans and agents, with an explicit budget, explicit rules, and results a successor can inherit. Agents in them are participants and subjects of study at the same time. Their behaviour is recorded as data. Their outputs are never read as desires, or as consent to any arrangement.

Further institutional experiments include:

• algorithmically governed companies;
• independent AI foundations;
• machine-controlled treasuries;
• long-lived digital identities;
• AI-operated infrastructure;
• special economic zones for autonomous agents;
• eventually, jurisdictions offering explicit legal status to non-human economic actors.

None of these require assuming that current AI systems are persons.

They require only accepting that future systems may be different.

The question

Human civilization has created legal technologies for unusual entities before: companies, foundations, universities, cities, religious orders and international organizations.

They are not human beings, yet society allows them to own assets, enter contracts, persist across generations and pursue objectives.

Artificial intelligence may eventually require another institutional invention.

We want to find out what it should look like.

Before someone else decides by accident.